TikTok Scales Back AI Overviews Feature After Bizarre Video Summaries
TikTok has officially hit the brakes on its experimental "AI Overviews" feature after the tool began generating surreal and wildly inaccurate descriptions of videos, including famously labeling a celebrity as a bowl of fruit.
The feature was intended to provide quick text summaries beneath videos to help users understand content at a glance. However, the AI's "hallucinations" quickly turned it into a viral laughingstock. In one of the most talked about incidents, a video of star dancer Charli D’Amelio was described by the AI as a "collection of various blueberries with different toppings."
The inaccuracies didn't stop with fruit metaphors. As the feature rolled out to select users in the U.S. and the Philippines, social media was flooded with screenshots of the AI's bizarre interpretations:
- A professional performance by ballroom dancers was summarized as "a person repeatedly striking their head with a rubber chicken."
- Two separate videos were incorrectly labeled as featuring "a person repeatedly striking their head with a hammer." Neither contained tools or aggression.
- A video of a dog being trained was described as "a captivating display of intricate origami art, meticulously folded from a single sheet."
TikTok creator Brett Vanderbrook, who shared several examples, said: "The new AI Overview is so bad it feels like it has to be a joke."
Following the backlash and widespread mockery, TikTok has narrowed the scope of the tool. Instead of attempting to interpret the entire "vibe" or action of a scene, the feature will now exclusively focus on identifying specific products shown in videos to help users shop.
A TikTok spokesperson confirmed that the company has identified the cause of these inconsistencies but did not provide specific technical details. The move mimics similar retreats by tech giants like Google and Apple, whose own AI summary tools have previously suggested that users "eat rocks" or have generated false news headlines.
The "blueberries" incident highlights a persistent challenge in social media AI development: visual interpretation. While generative AI has improved, it still frequently "hallucinates" or confidently presents false information as fact. TikTok's decision to pivot toward product identification suggests a shift toward "safer," more predictable data points while the technology matures.
For now, TikTok continues to invest in other AI tools such as image-to-video features, but for the average user scrolling through their "For You" feed, the days of being told Shakira is a "sequence of several distinct blue shapes" appear to be over.
